On Thu, 29 Apr 1999, Peter M. Saama wrote:
> Hello:
Dear Peter,
>
> When we fit spl(t), and dev(day) in ASREML, the .asr file contains output
> that looks like this:
>
> Source Model terms Gamma Component
> Compnt/StndErr
> spl(time) 18 18 6.65133
> 4.113614 1.41 P
> dev(time) 20 20 0.603323E-01
> 0.103056E-02 0.47 P
> ...
>
> In the above table is 4.113614 a smoothing parameter or is it a variance
> component? Furthermore, in the above table, is 0.103056E-02 a parameter
> used in specification of the roughness penalty or is it a variance component?
The component for spl(time) is a variance component, but it controls the
amount of smoothing. If this component (actually the ratio gamma) tends
to zero, the spline fit tends to a straight line, as it gets large the
fit tends to interpolation. The additional variance component, ie
dev(time), allows for the presence of group effects, and can only be
fitted if there is replication at each (or most) time point. This is in
line with the usual one-way random effects analysis where you would fit
y~ mu !r group
in asreml to get correct analysis ie estimating both an error and between
groups variance.
The usual roughness penalty that is described in most smoothing texts is
actually lambda = 1/gamma(spl).
I hope this helps
best wishes
> > Respectfully,
>
> Peter M. Saama /
> --
> Asreml mailinglist archive: http://www.chiswick.anprod.csiro.au/lists/asreml
>
...............................................................................
Brian Cullis Tel: 02 6938 1855
NSW Agriculture Fax: 02 6938 1809
Wagga Agricultural Institute email: brian.cullis@agric.nsw.gov.au
Pine Gully Rd
WAGGA WAGGA NSW 2650
--
Asreml mailinglist archive: http://www.chiswick.anprod.csiro.au/lists/asreml